Vol 3 (1) (2022) xxx-xxx

Measurement, control and automation

Website: https:// mca-journal.org

ISSN 1859-0551

Nonlinear control of a 3-DOF robotic arm driven by

electro-pneumatic servo systems
Duc Thinh Le1 , Van Trong Dang1 , Viet Hung Pham1 , Van Anh Nguyen Thi1 , Danh Huy Nguyen1 , Duc Truong Vu2 and
Tung Lam Nguyen1
1 Hanoi
University of Science and Technology
2 Le
Quy Don Technical University
* Corresponding author E-mails: huy.nguyendanh@hust.edu.vn and lam.nguyentung@hust.edu.vn

Abstract

The paper proposes an effective control approach to a 3-DOF serial robotic arm actuated by electro-pneumatic servo systems (EPSS). The
robot control problem is divided into robot dynamics and pneumatic actuators sub-systems, and the control is subsequently elaborated for
each sub-system. The challenging task arises when dealing with pneumatic systems where system parametric information is very difficult to
acquired correctly. Hence, the Lyapunov-based control is formulated to derive the control signal in presence of external disturbances and model
uncertainties thank to the integrated disturbance observer. The closed-loop system stability is analytically proven through a series of Lyapunov
candidate functions. Simulations using MATLAB/Simscape Multibody Link plug-in on Solidworks are carried out to demonstrate the control
validity.

Keywords: Backstepping Sliding Mode Control (BSMC), 3-DOF robotic arm, electro-pneumatic servo system, nonlinear disturbance observer.

Symbols Tóm tắt

Symbols Units Description Bài báo trình bày phương án điều khiển robot 3 bậc tự do được truyền
m1 kg Upper arm mass with cylinder động bằng khí nén. Bài toán điều khiển được chia nhỏ thành hai bài
m2 kg Forearm mass toán con bao gồm động lực học robot và cơ cấu chấp hành khí nén,
mf kg Load mass with translation joint cylinder dựa vào đó, bộ điều khiển được thiết kế trên các hệ con. Những thách
I1 kg · m2 Moment of inertia for upper arm thức khi điều khiển đến từ việc khó khăn trong việc đo lường, xác
around the shoulder định tham số của hệ thống truyền động khí nén. Trong bài báo này,
I2 kg · m2 Moment of inertia for forearm bộ điều khiển được thiết kế dựa trên lý thuyết ổn định Lyapunov có
around the elbow tích hợp bộ quan sát nhiễu dùng để ước lượng tham số bất định cũng
P1 P2 m Upper arm length như nhiễu tác động lên đối tượng. Tính ổn định của hệ thống kín được
P1 Pm1 m Distance from centroid of upper arm chứng minh và các kết quả mô phỏng cho thấy hiệu quả của phương
to shoulder pháp đề xuất.
P2 Pm2 m Distance from centroid of forearm
to elbow 1. Introduction
θ1 rad Upper arm rotation range
θ2 rad Forearm rotation range In recent years, advances in technologies facilitate many indus-
trial automation processes. This creates an increasing demand
for actuators driven in high speed for the purpose of reducing
Abbreviations time consumption in industrial tasks like assembly process. In
comparison with servo motors, electro-pneumatic servo sys-
PRA Pneumatic Robotic Arm tems (EPSS) offer better characteristics such as low cost, high
EPSS Electro-Pneumatic Servo System power-to-weight ratio, the absence of magnetic field and simple
DOF Degree of Freedom cleaning and maintenance. These distinctive advantages allow
NDO Nonlinear Disturbance Observer pneumatic-driven units to be integrated in a wide range of in-
dustrial actuators. Therefore, the problems of exact control of
this kind of actuators are necessary. However, many hindrances

Received: 13 December 2021; Accepted: 27 April 2022.

52
in model analyzing and controlling are found such as the com-
prehensibility of air and adverse effects of driving pneumatic
components in high speed that are oscillatory behaviours and
parasitic moving frictions [1].
The past decades also witnessed the profound automation in
many industries by the popularity of manipulators. Indeed, the
use of industrial manipulators improves the overall productivity
and reduces human error, as they have a superior load-lifting
capability and can replace human labor in dangerous environ-
ments. They perform procedures repeatedly with high precision
and are widely used from welding robots to car assembly lines.
Generally, manipulator joints are driven in different modes,
i.e., full-actuated [2] and under-actuated [3]. It is more chal-
lenging to do the under-actuated ones, since both kinematic
and dynamic constraints are tied in an under-actuated motion
system. Therefore, it is inappropriate to apply directly con-
ventional Euler–Lagrange method to this kind of joints. Thus,
complexity [4] and uncertainty of the model also degrade the
controller robustness. One typical model of a manipulator is
from [5, 6]. It is a 2-DOF robotic manipulator is comprised of
an upper arm, a forearm, a disc load, and a fixed torso. The
shoulder and elbow joints are driven to rotate by two Electro-
hydraulic actuators (EHAs). In this paper, we propose a model
using EPSS with additional electro-pneumatic servo actuator
at the end of the forearm, which governs the extension of the
forearm and makes our model 3-DOF, providing a better range
of work and applications. The 3D model of this manipulator is
illustrated in Figure 1.
Due to the enormous applicability of pneumatic units in in-
dustries, researchers have come up with different strategies to
control accurately this type of actuator [7]. An active rejection-
disturbance controller (ARDC) was utilized in [8] and [9] and
brought promising results. Pulse-width modulation technique
and fuzzy logic controller were combined in [10] to control
an electro-pneumatic servo systems. In [11], Karpenko et al.
proposed a quantitative feedback theory of a PI controller for
positioning of a servo-pneumatic servo actuator. Paper [12]
presented a hybrid controller of fuzzy controller and PID con-
troller. These linear control schemes experienced difficulties
in obtaining both fast response and precise position control at
the same time, due to the nonlinear nature of the EPSS.
Therefore, different approaches are implemented to tackle this
problem. Popular employed methods can be listed as backstep-
ping technique and Sliding Mode Control (SMC). Tracking
position control of EPSS is concerned in [13] with a backstep-
ping controller. In [14], authors applied adaptive backstepping
controller using Nussbaum gain with unknown model. A differ-
ent approach of adaptive backstepping control is implemented
in [15]. For SMC, it showed a wider range of application in
a lot of control scenarios: SMC for servo-pneumatic servo
system [16] and [17], SMC with friction compensation [18],
SMC with Particle Swarm Optimization [19], and adaptive
SMC [20]. These papers proved the robustness and efficiency
of SMC for a highly non-linear model like pneumatic system.
Some authors also combined these techniques together and ob-
tained even remarkably better control quality like Ren et al. [21]
and Lu et al. [22]. SMC algorithms in these mentioned papers
were found to be effective against external disturbances with
addition of an observer like extended state observer (ESO).
In this paper, a new control approach using Backstepping-
Sliding Mode Control (BSMC) and nonlinear high-gain dis-
Measurement, control and automation
turbance observer is presented to the 3-DOF pneumatic-driven
manipulator. BSMC is used in control the dynamics of the
EPAs and the kinematics of the entire manipulator itself. The
manipulator model under the effect of external disturbances
are also considered. The novel contributions of this paper are
summarized as follows:
(i). The 3-DOF robotic arm driven by EPSS including shoul-
der, elbow, and translational joint is built, in which the 3-DOF
manipulator is designed in CAD and a 3D model with precise
parameters of the model is built in Solidworks and Simscape
Multibody to validate the model in 3D simulation.
(ii). The nonlinear high-gain disturbance observer is embed-
ded with BSMC to have robust property against external dis-
turbances affecting on the manipulator. The stability of the
controller is proved mathematically by Lyapunov stability the-
ory. The performance of the BSMC with nonlinear high-gain
disturbance observer is compared to the conventional BSMC
using saturation function.
The reminder of this article is organized as follows: the Section
1 is introduction, and the Section 2 concerns about mathemat-
ical model of the manipulator. SMC with nonliear high-gain
DOB is proposed in Section 3, followed by the simulation eval-
uation in Section 4. Finally, Section 5 discusses the simulation
results and future development.
2. Mathematical modeling of the 3-DOF PRA
2.1. Dynamic and kinematic model of robotic arm
Figure 1: The 3-DOF pneumatic robotic arm configuration.
The 3-DOF pneumatic robotic arm is shown in Figure 1, in
which the shoulder, elbow, and translational joints are driven
by three independent single-rod cylinder electro-pneumatic
servo actuators. The robot position vectors Pm1 , Pm2 , and P3 are
given by, respectively:
 
−P1 Pm1 sin (θ1 + εm1 )
rPm1/O = (1)
P1 Pm1 cos (θ1 + εm1 )
 
−P1 P2 sin (θ1 ) − P2 P2m sin (θ1 + θ2 )
rPm2/O = (2)
P1 P2 cos (θ1 ) + P2 P2m cos (θ1 + θ2 )
Measurement, control and automation 53

where T1 = F1 L1 , T2 = F2 L2 , T3 = F3 ; L1 , L2 are the two dy-

  namic force arms, given as:
−P1 P2 sin (θ1 ) − z3 sin (θ1 + θ2 )
rP3 = (3)
P1 P2 cos (θ1 ) + z3 cos (θ1 + θ2 ) a1 + c1 2 − b1 2
  2 
L1 = a1 sin arccos
It is straightforward to determine the velocity vectors Pm1 , Pm2 , 2a1 c1
(12)
a2 + c2 2 − b2 2
  2 
and P3 as follows:
L2 = a2 sin arccos
  2a2 c2
−P1 Pm1 θ̇1 C (θ1 + εm1 )
vPm1/O = (4) Choosing the generalized coordinates vector q:
−P1 Pm1 θ̇1 S (θ1 + εm1 )
q = [θ1 θ2 z3 ]T (13)


−P1 P2 θ̇1 C (θ1 )−P2 P2m θ̇1 + θ̇2
C (θ1 + θ2 ) The Lagrange equation of the three-DOF robotic arm system
vPm2/O =
−P1 P2 θ̇1 S (θ1 )−P2 P2m θ̇1 + θ̇2 S (θ1 + θ2 ) is computed as:
(5)
∂ ∂E ∂E
T= − (14)
∂t ∂ q̇ ∂ q

vP3 = Applying the Lagrange equation, then the kinetic equation of


 the three-link robotic arm system is obtained as follows:
P1 P2 θ̇1 C (θ1 ) − z3 θ̇1 + θ̇2
C (θ1 + θ2 )−ż3 S (θ1 + θ2 )
−P1 P2 θ̇1 S (θ1 ) − z3 θ̇1 + θ̇2 C (θ1 + θ2 ) + ż3 C (θ1 + θ2 ) M (q, q̇) q̈ + C (q, q̇) q̇ + G = T (15)
(6)
where
where S (·) and C (·) denote sin (·) and cos (·), respectively.
 
M11 M12 M13
From the system representation, the total kinetic energy K of M (q, q̇) =  M21 M22 M23 
the 3-DOF PRA including the cylinders and the load mass is M31 M32 M33
computed as follows:  
C11 C12 C13
K = K1 + K2 C (q, q̇) =  C21 C22 C23 
1
2 1 1 C31 C32 C33
= m1 P1 Pm1 θ̇1 + I1 θ̇12 + m2 vPm2/O T vPm2/O (7)
2 2 2 
G1

1
2 1 T
+ I2 f θ̇1 + θ̇2 + m f vm f vm f G =  G2 
2 2 G3
where K1 and K2 denote the kinetic energy of the the upper arm
M11 = I1 + I2 + m1 P1 P1m 2 + m2 P2 Pm20 2 + m f x32
and the forearm, respectively. Similarly, the total gravitational
potential energy of the 3-DOF PRA including the cylinders + m2 P1 P2 2 + m f P1 P2 2 + 2m2 P1 P2 .P2 Pm20 cos θ2
and the load mass is computed by: + 2m f P1 P2 .x3 cos θ2
U = U1 +U2 +U f M12 = M21
= m1 gP1 Pm1 cos (θ1 + εm1 ) = I2 + m2 P2 Pm20 2 + m f x32 + m2 P1 P2 .P2 Pm20 cos θ2
(8)
+ m2 g (P1 P2 cos (θ1 ) + P2 P2m cos (θ1 + θ2 )) + m f P1 P2 .x3 cos θ2
+ m f g (P1 P2 cos (θ1 ) + z3 cos (θ1 + θ2 )) M13 = M31
= m f P1 P2 sin θ2
where U1 ,U2 , and U f are the gravitational potential energy
of the upper arm, the forearm, and the translational joint. g M22 = I2 + m2 P2 Pm20 2 + m f x32
denotes the gravitational acceleration. Then, the total energy E M23 = M32 = 0
of the three-link robotic arm system is given by: M33 = m f
E = K1 + K2 −U1 −U2 −U f (9) C11 = 0

By using simple geometric calculations, three cylinder dynamic C12 = m1 P1 P2 .ẋ3 cos θ2 − 2m2 P1 P2 .P2 Pm20 θ̇1 sin θ2
lengths c1 , c2 , c3 are represented by: − m2 P1 P2 .P2 Pm20 θ̇2 sin θ2 − 2m f P1 P2 .x3 θ̇1 sin θ2
r π  − m f P1 P2 .x3 θ̇2 sin θ2
c1 = a1 2 + b1 2 − 2a1 b1 cos − θ1 + ε11 C13 = 2m f x3 θ̇1 + 2m f x3 θ̇2 + 2m f P1 P2 .θ̇1 cos θ2
2
q
2
(10) C21 = 0
c2 = a2 2 + b2 − 2a2 b2 cos (π − θ2 − ε21 )
C22 = −m2 P1 P2 .P2 Pm20 θ̇1 sin θ2 − m f P1 P2 .x3 θ̇1 sin θ2
c3 = x3
C23 = 2m f x3 θ̇1 + 2m f x3 θ̇2 + m f P1 P2 θ̇1 cos θ2
Define the generalized force vectors T including torques T1 , T2 C31 = 0
and force F3 as:
C32 = m f P1 P2 θ̇1 cos θ2
T = [T1 T2 T3 ]T (11) C33 = 0
54 Measurement, control and automation

Figure 2: The quasi-physical model of the 3DOF-PRA based on Simscape Multibody.

2.3. Dynamic model of the EPSS

G1 = −m1 gP1 Pm1 sin (θ1 + ε11 )
+ m2 g (P1 P2 sin (θ1 + ε11 ) + P2 Pm2 sin (θ1 + θ2 )) The quasi-physical model of the 3DOF-PRA based on Sim-
scape Multibody is shown in Fig. 2. The force acting on each
− m f g (P1 P2 sin (θ1 + ε11 ) + x3 sin (θ1 + θ2 ))
robot joint is driven by the pneumatic cylinder actuator. The
G2 = m2 g.P2 Pm2 sin (θ1 + θ2 ) − m f gx3 sin (θ1 + θ2 ) pneumatic cylinders used to drive the three joints of the robot
G3 = m f g cos (θ1 + θ2 ) are assumed to have exactly the same physical structure, thus
The matrix M (q, q̇) is positive definite and symmetric for all the dynamic model of cylinders are represented in general as
q. Define state variables for the robotic arm x1 = q , then shown in Figure 3.
Equation (15) can be rewritten as follows:

ẋ1 = x2
(16)
ẋ2 = M−1 (T − Cx2 − G)

Let g2 (q) = M−1 and f2 (q) = −M−1 (Cx2 + G), Equation

(16) is obtained:

ẋ1 = x2
(17)
ẋ2 = f2 (q) + g2 (q) T Figure 3: Schematic of the pneumatic cylinder.

2.2. Inverse kinematic model of 3-DOF robotic arm

The length of the cylinder c is calculated by:
The goal of the 3-DOF robotic arm system is to control the c = 3L/2 + x p (22)
position of the robot’s final action. In order to obtain this,
from the position of the last action, the value of 2 rotation where x p is the piston position, L denotes the piston stroke. The
angles θ1 , θ2 and the displacement z3 of three robotic joints are force of the piston, which is caused by the pressure difference
designed as follows. between two chambers of the cylinder, is given by:
The position vector of the robot’s last action is defined as: Fa = P1 A1 − P2 A2 − Patm (A1 − A2 ) (23)
 T
P = px py (18) where A1 and A2 denote the piston effective areas. Patm , P1 ,
and P2 are the absolute pressures of the ambience, the actua-
Then from Figure 1, it can be shown that :
tor chamber 1, and the chamber 2, respectively. The control
p2x + p2y = P1 P2 2 + z23 + 2P1 P2 z3 cos (θ2 ) , (19) valve in the EPSS is a 5/3-way proportional valve, thus the
relationship between the control signals Av,i (i = 1, 2) is:
Av,1 = −Av,2 = Av (24)
p2x + p2y − P1 P2 2 − z23
cos (θ2 ) = . (20) Mass flows entering and leaving of the two cylinder chambers
2P1 P2 z3
are expressed by:
The rotation angles θ1 and θ2 of the robotic arm are calculated
as follows: µ̇i (Pu , Pd ) = Av ψi (Pu , Pd )
 Cd PuC1 Pd

θ1 = ATAN2 (py , px ) − ATAN2 (z3 sin θ2 , P1 P2 + z3 cos θ2 )
√
T s
, Pu ≤ Pcr
(25)


= Av 2
θ2 = ATAN2 (sin θ2 , cos θ2 ) .
P
d −P
 Cd√PuTC1 1 − P1−P , PPdu > Pcr
 u cr
(21) cr
Measurement, control and automation 55

where 3. Proposed approach

s
(γ+1)(γ−1)
3.1. Nonlinear high-gain disturbance observer

γ 2 (26)
C1 =
R γ +1
Parameter uncertainties and unknown disturbances are always
present in the 3-DOF robotic arm system. Thus, to obtain high-
 precision control, the model of the robot (17) is added with the
Ψ (Ps , P1 ) , i f Av ≥ 0
Ψ1 = (27) disturbances d (q) as:
Ψ (P1 , Patm ) , i f Av < 0

ẋ1 = x2
. (37)
 ẋ2 = f2 (q) + g2 (q) T + d (q)
Ψ (P2 , Patm ) , i f Av ≥ 0
Ψ2 = (28)
Ψ (Ps , P2 ) , i f Av < 0 where d (q) denotes the disturbances of the robot model in-
cluding the model uncertainties f̄2 (q), ḡ2 (q) and external dis-
Pu , Pd , and Ps are the upstream pressure, the downstream pres- turbances D, yielding:
sure, and the supply pressure, T is the upstream temperature
of air, Pcr is the critical pressure ratio, γ is the ratio of specific d(q) = f̄2 (q) + ḡ2 (q)T + D
heats, Cd is the discharge coefficient, R is the gas constant. The
pressure dynamics of the cylinder can be given as: Assumption 1. The state variables q of the 3-DOF robotic arm
driven by EPSS are physically bounded i.e. |q|< qmax , where
γRT γP1 A1 qmax is constant. The unknown disturbance vector d varies
Ṗ1 = µ̇1 − α ẋ p (29)
V1 V1 slowly and is bounded, the disturbances satisfy:

|ḋ| ≤ dmax (38)

γRT γP2 A2
Ṗ2 = µ̇2 + α ẋ p (30) where dmax are the constraint constants. Define the estimated
V2 V2
disturbance d̂, then estimated error is given by:
The volumes of two cylinder chamber can be calculated by: T
d̃ = d˜1 d˜2 d˜3 = d − d̂

  (39)
L
V1 = V01 + A1 + xp
2 Dynamic equation of estimated disturbance is determined as
  (31) [24]:
L
V2 = V02 + A2 + xp
2 1
d̂˙ = ẋ2 − f2 (q) − g2 (q) T − d̂ (q)


(40)
ε
Control valve servo input u is given by:
The estimation error dynamics of the disturbance observer is
Av = wkv u (32) obtained as follows [25]:

where w denotes valve orifice area gradient, kv is valve spool 1

d̃˙ = ḋ − d̃ (41)
position gain. From Equation (22) to (32), the force dynamics ε
of the pneumatic piston is obtained as follows:
Then,

P1 A21 P2 A22
   
Ḟa = −αγ ẋ p − + γRT
A1 ψ1 A2 ψ2
− wkv u d̃ ≤ e−(1/ε)t d̃ (0) + ερ (t) (42)
V1 V2 V1 V2
(33) where ρ (t) denotes an envelope function that ρ (t) ≥ ḋ
Remark 1. Equation (42) shows that the estimated error d̃ of
Let the nonlinear high-gain disturbance observer can made arbitrar-
 T ily small when the observer gain ε is chosen as infinitesimal.
x= xp ẋ p (34)
3.2. Robust control design process
Then, Equation (33) can be rewritten as:
In the paper, the complete system including the robot and the
Ḟa = fa (x,t) + ga (x,t) u (35) EPSS is rendered as a cascade system. Firstly, the system
level is taken into account to identify required torque/force to
where achieve positioning tasks. Subsequently, the control design
is carried out at pneumatic actuator level to construct the
P1 A21 P2 A22
 
fa (x,t) = −αγ ẋ p − final control signal. In this step, the nonlinear high-gain
V1 V2 disturbance observer is designed to effectively estimate
  (36)
A1 ψ1 A2 ψ2 external disturbances acting on the system.
ga (x,t) = γRT − wkv
V1 V2
a. The 3-DOF robotic arm control
In this paper, we assume that all the EPSS are identical.
56 Measurement, control and automation

The conventional backstepping technique is integrated with the of the EPSS, sliding mode control is used for the electro-
sliding mode control to obtain a robust controller for uncer- pneumatic servo control system. Define sliding variable sa as
tain systems. The basic BSMC for the 3-DOF robotic arm is follows:
designed by firstly defining the tracking error as:
sa = Fa − Fd (53)
e1 = x1 − x1d (43)
where Fd denotes the desired force that determined from the
where x1d denotes the desired trajectory. Select Lyapunov can- control laws Td via the generalized force vectors (11) and
didate function V1 as: the two dynamic force arms (12). Select Lyapunov candidate
function V3 as:
1
V1 = eT1 e1 (44) 1
2 V3 = sTa sa (54)
2
The time derivative of Equation (44) is taken as follows:
The time derivative of equation (54) is taken as:
V̇1 = eT1 ė1 = eT1 (ẋ2 − ẋ1d ) (45) V̇3 = sa T ṡa = sa fa + ga u − Ḟd


(55)
Equation (45) suggests that the virtual control law is chosen Based on sliding mode control, the control law u of electro-
as: pneumatic servo system is designed as follows:
x2d = −c1 e1 + ẋ1d u = ga −1 −fa − ksa + Ḟd − κsgn (sa )


(46) (56)

where c1 = diag(c11 , c12 , c13 ) is a positive definite matrix, then, where k = diag(k1 , k2 , k3 ) and κ = diag(κ1 , κ2 , κ3 ) are positive
V̇1 ≤ 0. Define a sliding variable s = [s1 s2 s3 ]T : definite matrices.

s = x2 − x2d (47) 3.3. Stability analysis

Subsequently, Lyapunov candidate function V2 is defined as Theorem 1. Consider the 3-DOF robotic arm driven by
follows: electro-pneumatic servo systems described in general form as
1 equations (17) and (35) under unknown disturbance bounded
V2 = V1 + sT s (48) as Assumption 1, the control laws of the robotic arm (51),
2
the control laws of the EPSS (56), and the observer gains
The time derivative of equation (48) is taken as ε of the nonlinear disturbance observer (40) guarantee the
Input-to-State Stability [23] of the control system.
V̇2 = V̇1 + sT ṡ
(49)
= −c1 eT1 e1 + s (f2 (q) + g2 (q) T + d (q) − ẋ2d ) Proof of Theorem 1. Lyapunov function candidate is defined
as:
Equation (49) directly indicate that control law T of the 3-DOF 1 1 1 1
robotic arm is designed as: V = eT1 e1 + sT s + d̃T d̃ + sTa sa (57)
2 2 2 2
Td = g2 −1 (q) (−f2 (q) − c2 s + ẋ2d − υsgn (s)) (50) The time derivative of V is taken as:

where c2 = diag(c21 , c22 , c23 ) and υ = diag(υ1 , υ2 , υ3 ) are V̇ = e1 T ė1 + d̃T d̃˙ + sT ṡ + sa T ṡa (58)
positive definite matrices. Using the estimated disturbance
signal (40), Equation (50) is rewritten as follows: Substituting the control laws (51), (56) and the estimation error
dynamic (41) into Equation (58), it can be rewritten as:
Td = g2 −1 (q) −f2 (q) − c2 s + ẋ2d − υsgn (s) − d̂ (q)


(51)
V̇ = −eT1 c1 e1 + sT −c2 s − υsgn (s) + d − d̂

(59)
 
The control design for the robot level is completed, in the next 1
+d̃ ḋ − d̃ + sa T (−ksa − κsgn (sa ))
T
section, the control action for the EPSS is established. ε

b. The electro-pneumatic servo system control The inequality |a||b|⩾ ab is used for Equation (59), thus we
The dynamic model of one cylinder is described in detail in obtain:
subsection 2.3. To calculate the control for the three-cylinder of 
si ζi ˜ 2

T T
the robot system, then the equation (35) is rewritten as follows: V̇ ≤−e1 c1 e1 − sa ksa − c2i − di
ζi 2c2i
2 !
ηi 2 2

1 1
Ḟa = fa + ga u (52) −τi |d˜i |− |d˙i | − 1 − 2 c2i si 2 − d̃i (60)
2τi ζi 4c2i
T  
where Fa =

Fa,1 Fa,2 Fa,3 , fa = 1 ˙2
+ |di |
 T 
fa,1 (x,t) fa,2 (x,t) fa,3 (x,t) , u = u1 u2 u3
T
, 4τi
and ga = diag (ga,1 (x,t), ga,2 (x,t), ga,3 (x,t)), the subscript where
n = [1, 2, 3] represent the cylinders of the three joints (i.e, !
shoulder, elbow, and translation joint), respectively. 1 ζi 2 + ηi 2
τi = − ;ζ > 1; η ̸= 0; i = 1, 2, 3 (61)
In order to reduce the influence of parameter uncertainties ε 4c2i
Measurement, control and automation 57

Figure 4: The EPSS actuated 3-DOF robot control structure.

There exists constants ι and γ satisfying the following condi- Table 1: Parameters of the 3-DOF robotic arm and the proposed
tions: controller

3   Symbol Value Unit
ι = ∑ 1 |d˙i |2 ,

4τi m1 1.335 kg
i=1 n  o (62)
γ = min 1 − 12 c2i , ci , ki , ηi 2
 m2 0.86 kg
ζi 4c2i
mf 1 kg
I1 0.05628 kg · m2
Equation (62) can be rewritten as:
I2 0.21154 kg · m2
V̇ ⩽ −2γV + ι (63) P1 P2 0.3 m
P1 Pm1 0.1523 m
Consequently, it can be concluded that P2 Pm2 0.165 m
L 0.2 m
ι
V (t) ⩽ V (0) e−2γt + 1 − e−2γt


(64) Patm 0.1 × 106 Pa
2γ Ps 0.6 × 106 Pa
Remark 2. Equation (64) implies that the tracking errors e1 , s, Pcr 0.528
sa , and estimation error d̃ of the 3-DOF robotic arm driven γ 1.4
by electro-pneumatic servo systems using the backstepping- R 287 N · m/kg · K
sliding mode control with nonlinear high-gain disturbance ob- T 300 K
server exponentially converge to an arbitrarily tiny ball. c1 = diag(c11 , c12 , c13 ) diag(2.5, 2, 4)
Remark 3. In order to reduce the chattering effect of sliding c2 = diag(c21 , c22 , c23 ) diag(5, 7, 13)
mode control, the saturation function is used to replace the sign υ = diag(υ1 , υ2 , υ3 ) diag(1, 2, 1)
function in the control laws (51), (56). ε 0.015

4. Simulation evaluation
Simulations based on Matlab/Simscape Multibody Link plug-
in on Solidworks are carried out to demonstrate the control
validity of the 3-DOF robotic arm driven by electro-pneumatic
servo systems. In which, the electro-pneumatic servo systems
are controlled by sliding mode control, and backstepping-
sliding mode control with nonlinear high-gain disturbance
observer is used for the 3-DOF robotic arm that are shown
as in Figure 4. In the simulations, assume that disturbances
acting on robot joints is sinusoidal wave form of 0.5 sin πt as
shown in Figure 11. Parameters of the 3-DOF robotic arm and
the proposed controller is expressed in Table 1. Figure 5: Shoulder joint rotation angle.
From figures 5, 6, and 7, we can observe that angular and
translational motions of the robot under the proposed con-
troller integrated with the disturbaces observer exhibit better is presented in Figure 11, in which the disturbance estima-
performances compared to traditional backstepping-sliding tion of the three joints (i.e, shoulder, elbow, and translational
mode control. Figures 8, 9, and 10 show that forces produced joint) follows the disturbances of 3-DOF robotic arm. By us-
by the EPSS closely (,i.e. the actual forces) track the desired ing the nonlinear high-gain disturbance observer combined
forces determined by the robot motion control problem. It is with backstepping-sliding mode control, the rotation angle re-
also noted that the control forces are well maintained in ap- sponses of the shoulder and elbow joints, and translational
plicable ranges. The effectiveness of the disturbance observer joint displacement are greatly improved that reducing the in-
58 Measurement, control and automation

Figure 6: Elbow joint rotation angle. Figure 10: Force response of the pneumatic cylinder F3 .

Figure 11: Estimated disturbance responses of robot joints.

Figure 7: Translational joint displacement.

5. Conclusion
In the paper, the backstepping-sliding mode control with em-
bedded disturbance is developed for the 3-DOF robot driven by
electro-pneumatic actuators. The complicated dynamics of the
electro-pneumatic characteristic is considered in the control
design. By backstepping-like procedure the control scheme
is formulated exhibiting robust property against disturbances.
The constructed control allows the system output track the de-
sired signal in a short time period with negligible steady state
errors. Currently, the simulation is studied in the paper. In the
near future, the experimental results will be presented.

Figure 8: Force response of the pneumatic cylinder F1 . Acknowledgement

Van Trong Dang was funded by Vingroup JSC and supported
by the Master, PhD Scholarship Programme of Vingroup In-
novation Foundation (VINIF), Institute of Big Data, code
VINIF.2021.ThS.26.

References
[1] Kawakami, Y., Moguchi, H., & Kawai, S. (1990) Some considerations
on the high-speed driving of pneumatic cylinders. J. Japan Hydraulics
& Pneumatics Society, 21(3 May), 318–327.
[2] Zhijun Li, Shuzhi Sam Ge, Adams, M., & Wijesoma, W. S. (2008)
Adaptive Robust Output-Feedback Motion/Force Control of Electrically
Driven Nonholonomic Mobile Manipulators. IEEE Transactions on
Control Systems Technology, 16(6), 1308-1315.
[3] Dong, W. (2002) On trajectory and force tracking control of constrained
mobile manipulators with parameter uncertainty. Automatica, 38(9),
Figure 9: Force response of the pneumatic cylinder F2 . 1475–1484.
[4] Yan, Y., Xu, J., & Wiercigroch, M. (2017) Basins of attraction of the
bistable region of time-delayed cutting dynamics. Physical Review E,
96(3).
[5] Guo, Q., Zhang, Y., Celler, B. G., & Su, S. W. (2019) Neural Adap-
tive Backstepping Control of a Robotic Manipulator with Prescribed
fluence of the disturbance of the three joints compared to the Performance Constraint. IEEE Transactions on Neural Networks and
conventional backstepping-sliding mode control. Learning Systems, 30(12), 3572–3583.
Measurement, control and automation 59

[6] Guo, Q. & Jiang, D. Nonlinear Control Techniques for Electro-

Hydraulic Actuators in Robotics Engineering. (CRC Press,2017)
[7] Ali, H., Bahari, S., Noor, M.S., Bashi, S.M., & Marhaban, M.H. (2009)
A Review of Pneumatic Actuators (Modeling and Control). Australian
Journal of Basic and Applied Sciences, 3 ,pp. 440 - 454.
[8] Wang, Y.-B & Bao, G. & Wang, Z.-W. (2007) Application of auto-
disturbance rejection controller for pneumatic servo system. Journal of
Dalian Maritime University, Vol 33.
[9] Zhao, L., Zhang, B., Yang, H., & Li, Q. (2017) Optimized linear active
disturbance rejection control for pneumatic servo systems via least
squares support vector machine. Neurocomputing, 242, 178–186.
[10] Najjari, B., Barakati, S. M., Mohammadi, A., Futohi, M. J., & Bostanian,
M. (2014) Position control of an electro-pneumatic system based on
PWM technique and FLC. ISA Transactions, 53(2), 647–657.
[11] Karpenko, M., & Sepehri, N. (2004) QFT design of a PI controller
with dynamic pressure feedback for positioning a pneumatic actuator.
Proceedings of the 2004 American Control Conference.
[12] Parnichkun, M., & Ngaecharoenkul, C. (2001) Kinematics control
of a pneumatic system by hybrid fuzzy PID. Mechatronics, 11(8),
1001–1023.
[13] Smaoui, M., Brun, X., & Thomasset, D. (2006) A study on tracking
position control of an electropneumatic system using backstepping
design. Control Engineering Practice, 14(8), 923–933.
[14] Ren, H.-P., Wang, X., Fan, J.-T., & Kaynak, O. (2019) Adaptive Back-
stepping Control of a Pneumatic System With Unknown Model Parame-
ters and Control Direction. IEEE Access, 7, 64471–64482.
[15] Ren, H.-P., & Huang, C. (2013)) Adaptive backstepping control of pneu-
matic servo system. 2013 IEEE International Symposium on Industrial
Electronics.
[16] Carneiro, J. F., & de Almeida, F. G. (2014) Accurate motion control
of a servopneumatic system using integral sliding mode control. The
International Journal of Advanced Manufacturing Technology, 77(9-
12), 1533–1548.
[17] Blagojević, V., Šešlija, D., Stojiljković, M. & Dudić, S. Efficient control
of servo pneumatic actuator system utilizing by-pass valve and digital
sliding mode. Sadhana. 38, 187-197 (2013)
[18] Tran, X., Nguyen, V., Nguyen, N., Pham, D. & Phan, V. Sliding mode
control for a pneumatic servo system with friction compensation. Inter-
national Conference On Engineering Research And Applications. pp.
648-656 (2019)
[19] Liu, G., Li, G., Song, H. & Peng, Z. Neural network sliding mode
control for pneumatic servo system based on particle swarm optimiza-
tion. Proceedings Of The 11th International Conference On Modelling,
Identification And Control (ICMIC2019). pp. 1239-1248 (2020)
[20] Plestan, F., Shtessel, Y., Brégeault, V., & Poznyak, A. (2010) New
methodologies for adaptive sliding mode control. International Journal
of Control, 83(9), 1907–1919.
[21] Ren, H., & Fan, J. (2016) Adaptive backstepping slide mode control
of pneumatic position servo system. Chinese Journal of Mechanical
Engineering, 29(5), 1003–1009.
[22] Lu, C.-H., Hwang, Y.-R., & Shen, Y.-T. (2010) Backstepping Sliding-
Mode Control for a Pneumatic Control System. Proceedings of the
Institution of Mechanical Engineers, Part I: Journal of Systems and
Control Engineering, 224(6), 763–770.
[23] Sontag, E. Input to state stability: Basic concepts and results. Nonlinear
And Optimal Control Theory. pp. 163-220 (2008)
[24] Dang, V., Le, D., Nguyen-Thi, V., Nguyen, D., Tong, T., Nguyen, D.,
Nguyen, T. & Others Adaptive Control for Multi-Shaft with Web Mate-
rials Linkage Systems. Inventions. 6, 76 (2021)
[25] Le, D., Nguyen, D., Le, N. & Nguyen, T. Traction control based on
wheel slip tracking of a quarter-vehicle model with high-gain observers.
International Journal Of Dynamics And Control. pp. 1-8 (2021)